For engineers who ship open source with coding agents

We pay open-source engineers for the work they already do.

Your traces become RL environments in real time. You get a share of every sale.

curl -fsSL https://benchturn.com/install.sh | sh

Keys and .env files are never recorded. Any repo with a .gen-ignore file is skipped.

macOS or Linux. Takes about a minute. Read the script.


            
            
          

Records the coding agents you already use

  • Claude Code
  • Codex
  • Cursor
  • Gemini CLI
  • GitHub Copilot CLI
  • Cline
  • Pi
  • OpenCode
// how it works

Install once. Keep working. Get paid.

There are no tasks to pick up and nothing to annotate. benchturn records your coding-agent sessions on open-source projects, turns the good ones into environments, licenses them to labs, and pays you a share of every sale.

01

One command

One self-contained binary. benchturn init signs you in with GitHub and connects every coding agent it finds.

$ benchturn init ✓ Signed in as octocat ✓ Recording Claude Code, Codex and Cursor.
02

Work like you do now

Every prompt in a git repo captures the tree before and after, plus the transcript. Uploads run on a timer; your agent never slows down.

$ benchturn status captured tokio-rs/tokio queued upload in background
03

Get a share of the sale

Good sessions become environments labs license. When one sells, your share lands on your portal and pays out monthly.

$ benchturn account opens your portal: sessions, states, earnings
// what gets recorded

Exactly what a lab needs. Nothing it doesn't.

A real repository at a real commit, a task someone actually asked for, and a way to check the result. Work outside a git repository is not recorded.

per prompt, inside a git repo recorded
  • The repository tree before the prompt and after the agent stops, so the change can be replayed exactly.
  • The transcript of that turn: your prompt, the agent's messages, and the tools it called.
  • The model name and a timestamp. At the end of a session, the full transcript.
Captures live under ~/.claude/gen until uploaded, and a copy stays there.
ever, anywhere never
  • Keys, tokens, passwords and .env files. A secret scan runs on your machine before upload and again at ingest. A capture that trips it is dropped.
  • Your name. A lab receives the public repository, the prompts and the transcript, with no account information attached.
  • Excluded repositories. Anything in a repo with a .gen-ignore file never leaves your machine.
Private repositories can be recorded for you, but only sessions on public repositories become environments.
// where a session goes

From your terminal to a lab, one state at a time.

01
Received

Uploaded and on your portal.

02
Pending review

Real project, real code changed, a task that can be checked?

03
Environment built

Rebuilt as a reproducible task.

04
Listed

Offered to labs.

05
Licensed

A lab bought it. Your share is added to your pending payout.

A session that is not a fit is deleted. We email you when a state changes. We don't promise a date.

// past sessions count too

Your best work is already on disk.

Run benchturn past and your own agent reads the sessions already on your machine, picks the ones on public repositories with real work in them, checks each for private material, and shows you the list. Nothing is uploaded until you say yes to each one.

The instructions it follows are public at benchturn.com/agents.md.

// faq

Questions, answered.

Which agents does it support?

Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot CLI, Cline, Pi and OpenCode. Two need one manual step, which benchturn init prints: Codex asks you to approve the hooks once with /hooks, and Cline has a Hooks switch under Settings → Features.

Will it slow my agent down?

No. Uploads run in the background on a timer. Nothing waits on them.

What makes a session a good fit?

A real project that other people use, big enough to be hard, with tests or a way to check a change, that builds on Linux. And real work on it: a bug fixed, a feature added, or a refactor carried through. Best of all, you steered: corrected the agent, rejected its approach, or knew something it did not.

How do I stop recording a repository?

Run benchturn ignore inside it. That writes a .gen-ignore file and keeps it out of git; the repository is skipped entirely. benchturn pause stops all recording for a while.

How do I remove it?

benchturn uninstall takes the hooks out of every agent's settings and leaves your own agents' files exactly as they were. benchturn delete <id> removes one capture here and tells us to forget it.

Where can I read more?

Everything is in the docs. Or write to hello@benchturn.com.

Your next session could already be worth something.

macOS or Linux, with git. Takes about a minute.

curl -fsSL https://benchturn.com/install.sh | sh